I Gede Susrama Mas Diyasa
UPN Veteran Jawa Timur

Published : 3 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 3 Documents
Search

Image Synthesis for Sperm Dataset Augmentation using WGAN-GP Hajjar Ayu Cahyani Kuswardhani; I Gede Susrama Mas Diyasa; Mohammad Idhom
IJCONSIST JOURNALS Vol 5 No 2 (2024): March
Publisher : International Journal of Computer, Network Security and Information System

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33005/ijconsist.v5i2.146

Abstract

This research explores the efficacy and limitations of applying a Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) to generate synthetic human sperm microscopy images for data augmentation. We assessed the WGAN-GP's performance on a complex, heterogeneous dataset where images contained multiple object types. Despite achieving stable training convergence, the model's output quality was suboptimal, as evidenced by a high Fréchet Inception Distance (FID) score of 134 and qualitative signs of partial mode collapse. The generator struggled to capture the complete morphological diversity of the sperm cells. A second experiment using a dataset pre-sorted into distinct classes (Normal, Abnormal, Non-Sperm) yielded a marked improvement. This approach led to substantially lower FID scores (59.19, 74.92, and 83.56) and exhibited more robust training dynamics. Our findings underscore a critical conclusion: the success of WGAN-GP in this domain is fundamentally tied to the simplicity of the data distribution. We recommend that future efforts leverage class-conditioned models, simplified data structures, and refined generator architectures to achieve high-precision augmentation for medical imaging tasks.
Design and Construction of a Multifunctional Educational Mobile Robot for STEM Learning Based on Visual Programming Using ESP32 Mochamad Farid Fadhlulloh; I Gede Susrama Mas Diyasa; Basuki Rahmat
Journal of Artificial Intelligence and Engineering Applications (JAIEA) Vol. 6 No. 1 (2026): October 2026
Publisher : Yayasan Kita Menulis

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.59934/jaiea.v6i1.2637

Abstract

The limited availability of STEM learning media capable of interactively and affordably integrating cross-disciplinary concepts poses a significant barrier to STEM education implementation in Indonesian secondary schools. This study proposes and realizes a Multifunctional Educational Mobile Robot based on the ESP32 microcontroller that can be visually programmed using PictoBlox—a Scratch-based platform—as a solution to this gap. The system is designed to operate in four modes: Line Follower (TCRT5000 line sensor), Obstacle Avoidance (HC-SR04 ultrasonic sensor), analog joystick control, and Android smartphone control via Bluetooth using the Dabble application. An aluminum chassis with Mecanum wheels enables omnidirectional movement that enriches the coverage of vector physics concepts. Systematic testing yielded quantitative data: the HC-SR04 sensor achieved a stable error of 2.80% across the 10–80 cm; the TCRT5000 sensor achieved 96.7% accuracy under standard conditions; Bluetooth communication maintained a latency of 118–122 ms with zero packet loss at 1–8 meters; and the success rates for Line Follower, Obstacle Avoidance, and Smartphone Control modes were 93.3%, 96.0%, and 97.0%, respectively. The system demonstrated full reliability during continuous 30-minute sessions. Pedagogical analysis confirmed that the four modes collectively cover all STEM domains, making this robot a comprehensive STEM learning media candidate for senior and vocational high school levels.
Systematic Evaluation of YOLOv8 Training Configurations for Detecting Bovine Sperm Head Abnormalities Rizqy Khoirul Waritsin; I Gede Susrama Mas Diyasa; Eva Yulia Puspaningrum
Journal of Artificial Intelligence and Engineering Applications (JAIEA) Vol. 6 No. 1 (2026): October 2026
Publisher : Yayasan Kita Menulis

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.59934/jaiea.v6i1.2713

Abstract

Bovine sperm morphology assessment is an important stage in evaluating semen quality for artificial insemination programs. Manual examination under a microscope is time-consuming and highly dependent on the observer's expertise, motivating the development of automated detection systems using deep learning. This study systematically evaluates the influence of YOLOv8 training configurations for detecting bovine sperm head abnormalities. A total of 27 experimental scenarios were designed by combining three optimizers (SGD, Adam, and RMSProp), three learning rates (0.0001, 0.001, and 0.01), and three training epochs (75, 100, and 125). Contrast Limited Adaptive Histogram Equalization (CLAHE) was applied to improve image quality before model training. Performance was evaluated using Precision, Recall, F1-score, and mAP@0.5. The experimental results indicate that the training configuration significantly affects detection performance. The best performance was achieved using the RMSProp optimizer with a learning rate of 0.01 and 75 training epochs, producing a Precision of 89.81%, Recall of 88.38%, F1-score of 89.08%, and mAP@0.5 of 91.76%. These findings demonstrate that systematic evaluation of training configurations provides practical guidance for improving YOLOv8-based bovine sperm abnormality detection without modifying the network architecture